Full Terms & Conditions of access and use can be found athttps://www.tandfonline.com/action/journalInformation?journalCode=tgei20

Geocarto International

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tgei20

Inversion of soil roughness for estimating soilmoisture from time-series Sentinel-1 backscatterobservations over Yanco sites

Ju Hyoung Lee & Jeffrey Walker

To cite this article: Ju Hyoung Lee & Jeffrey Walker (2020): Inversion of soil roughness forestimating soil moisture from time-series Sentinel-1 backscatter observations over Yanco sites,Geocarto International, DOI: 10.1080/10106049.2020.1805030

To link to this article: https://doi.org/10.1080/10106049.2020.1805030

View supplementary material

Accepted author version posted online: 05Aug 2020.Published online: 27 Aug 2020.

Submit your article to this journal

Article views: 11

View related articles

View Crossmark data

Inversion of soil roughness for estimating soil moisturefrom time-series Sentinel-1 backscatter observations overYanco sites

Ju Hyoung Leea and Jeffrey Walkerb

aResearch Institute for Mega Construction, Korea University, Seoul, Republic of Korea; bDepartmentof Civil Engineering, Monash University, Clayton, Australia

ABSTRACTSentinel-1 provides improved temporal and spatial resolutions incomparison with previous satellite missions. Unlike the changedetection method that assumes the time-invariance of surfaceroughness at a coarse resolution, this study spatially invertedroughness and soil moisture from Sentinel-1 backscatter.Although it is important at high-resolution, local measurements ofsurface roughness are not available in an operational setting.Thus, question was how often time-varying soil roughness infor-mation should be updated in operations for reasonable soil mois-ture retrievals but at effective computational cost. Localvalidations show that a monthly update for surface roughness issufficient for soil moisture retrievals. In more details, Root MeanSquare Errors (RMSE) of 0.01m3/m3 is found at Yanco A3 site, and0.03m3/m3 at Yanco A5 site, with differences in backscatteringbetween Integral Equation Model (IEM) simulation and Sentinel-1measurement of 0.78 dB at Yanco A3 site and 0.012 dB at YancoA5 site, respectively.

ARTICLE HISTORYReceived 19 April 2020Accepted 23 July 2020

KEYWORDSSynthetic Aperture Radar(SAR) soil moisture;Sentinel-1; inversion ofroughness; small-scale roughness

1. Introduction

Soil moisture is an important variable in the water, energy and carbon cycles. It is regu-larly used in hydrologic, land surface and climate model initialization, as it is fundamen-tally involved in the energy and water balance as a controlling factor on rainfallinfiltration, energy flux, and deep drainage (Niu et al. 2011; Lee et al. 2014). The historicalrecords of soil moisture accumulated over several years provide an essential indicator ofclimate change, while real-time soil moisture can be used for predicting urgent disasterssuch as landslides, flooding, and fire events (Henry et al. 2006; Jensen et al. 2018).However, despite significance, it is difficult to accurately simulate or predict soil moisturewith models alone (Paloscia et al. 2013). That is because uncertainties arise from errors inrainfall, or soil hydraulic properties, and from the complex relationship between land sur-face state evolution and meteorological forcing (Taylor et al. 2012).

CONTACT Ju Hyoung Lee ju.lee@mail.com Research Institute for Mega Construction, Korea University, 145Anam-ro, Seongbuk-gu, Seoul, Republic of Korea.

Supplemental data for this article can be accessed at https://doi.org/10.1080/10106049.2020.1805030.

� 2020 Informa UK Limited, trading as Taylor & Francis Group

GEOCARTO INTERNATIONALhttps://doi.org/10.1080/10106049.2020.1805030

One way to globally estimate the spatial distribution of soil moisture and simulate sucha global circulation between land surface and meteorological forcing is to measure it withsatellites. The microwave sensors that are predominantly used for estimating soil moistureoperate at C- and L-bands. The first satellite that aimed for retrieving global soil moisturewas the ESA (European Space Agency)’s Soil Moisture Ocean Salinity (SMOS) missionlaunched in 2009 (Kerr et al. 2012). The National Aeronautics and Space Administration(NASA)’s Soil Moisture Active Passive (SMAP) mission was the next soil moisture satel-lite launched in 2015. Both are passive microwave satellites operating at L-band(1.413GHz, 21 cm, 2 to 3 day of revisit time). Their spatial resolutions are low at approxi-mately 40 km, although their products are provided on the grids of 25 km or 9 km. Theissue is that such low resolutions do not properly recognize the land surface heterogeneity(Kornelsen and Coulibaly 2013), which alter at the scale of meters.

In contrast, Sentinel-1, an active microwave sensor at C-band, provides high resolutiondata, although active microwave data are much more challenging to interpret. It is oftendiscussed that difficulty arises from the radar signal complicated by dense vegetation orinstrumental configuration. Sentinel-1A carries a Synthetic Aperture Radar (SAR) thatoperates at C-band (5.4GHz) with dual polarization (VVþVH, HHþHV), and fourexclusive acquisition modes of strip map, interferometric wide swath, extra-wide swath,and wave mode. Its incidence angle has a variation of 20� 46� (Bauer-Marschallingeret al. 2019). Its revisit time is 12 days, while it is 3 days at equator.

Various approaches are available to retrieve soil moisture from SAR measurements(Kornelsen and Coulibaly 2013). Nonetheless, it still remains a complicated and ill-posed problem, because backscattering is not only affected by dielectric constant (i.e.,soil moisture) but also by other variables such as soil roughness, volumetric scatteringfrom the soil and vegetation, and SAR configuration including incidence angle, wave-length/frequency of instrument and polarization (Mirsoleimani et al. 2019). Moreover,soil moisture retrieval has the nature of equi-finality (Beven and Freer 2001). Multiplephysical parameter combinations often produce the same SAR backscattering intensity.For example, wet soils produce a similarly high SAR backscatter intensity like roughsoils, dense vegetation or rocks. Thus, for a successful retrieval, it is required to makean appropriate estimation of other physical parameters in order to isolate the contribu-tion from soil moisture.

One of the common methods for retrieving soil moisture from active microwavesensors or scatterometers is change detection. It decomposes the radar backscatteringinto largely two simple elements: dielectric constant and roughness (Bauer-Marschallinger et al. 2019). In a dry state where the effects of dielectric constant arenegated, it is considered that the radar backscattering comes from the surface rough-ness and vegetation only. In contrast, backscattering for wet soils is interpreted as soilmoisture by subtracting backscattering in the dry conditions (i.e., roughness) . In otherwords, land surface homogeneity and time-invariance of surface parameters is assumedwithin low-resolution scatterometer footprint rather than estimating surface roughnessor vegetation parameters. Thus, backscatter data at a coarse resolution and high tem-poral sampling have shown interesting results with this approach. However, such achange detection is not appropriate for high-resolution SAR data over spatially hetero-geneous land undergoing vegetation distribution, tillage or soil erosion. Hence, withrespect to the applicability of change detection, Wagner et al. (1999, 2010) clarifiedthat to assume time-invariance of surface roughness is reasonable at low-resolutionscatterometer, but inapplicable to high-resolution. Sentinel-1 SAR data corresponds tothe latter.

2 J. H. LEE AND J. WALKER

One of the useful approaches to retrieve such heterogeneous land surface parameters athigh-resolution is an inversion. Unlike change detection, it characterizes the physicalparameters by minimizing the difference between model estimates and instrument meas-urements (Pauwels et al. 2002). This inversion approach is still useful, although otherapproaches such as fractal method or Artificial Neural Networks (ANNs) are suggested toestimate SAR roughness, recently. That’s because they all require some roughness inputsas a priori. For example, to generate fractal surface, spatially distributed root-mean-squareheights are needed. This SAR roughness is fundamentally different from roughness meas-ured in the field, due to difference in scale and various instrumental factors to have aninfluence on backscatter. In addition, when estimating SAR roughness or soil moisturewith ANNs, we still need to provide roughness inputs used as ‘target’ vectors (i.e., desiredoutcome). Thus, despite other alternatives, the need for inversion methods is very clear.

Several studies employed inversion methods for SAR retrievals. van der Velde et al.(2012) inverted soil roughness by comparing the ENVISAT Advanced Synthetic ApertureRadar (ASAR) backscattering measured from three different incidence angles with theIEM (Integral Equation Model) backscattering. They discussed that an enhanced temporalresolution is needed, suggesting to apply the approach for Sentinel-1 data as future works.Pierdicca et al. (2010) retrieved soil moisture content in a vegetated area, using multi-temporal Airborne SAR and biomass data from Landsat reflectance to correct the effectsof vegetation. After generating a probability distribution function of a target parameterconditioned to SAR backscattering, soil moisture was inverted from backscattering. Theyconcluded that this inversion approach better deals with ill-posed problems complicatedby vegetation.

For supporting an inversion, Look Up Table (LUT) has also been widely used(Rahman et al. 2007; Merzouki et al. 2011). Forward model outputs of backscattering areestablished for a diverse range of input parameters, for example, soil roughness condi-tions, dielectric constant or incidence angle. The physical parameters to produce the simi-lar backscattering coefficients with the backscattering measurements are selected fromthe LUT.

On the other hand, soil moisture can also be stochastically retrieved with a Bayesianapproach, which uses the probability distribution function (PDF) generated by a broadrange of dielectric constant and roughness conditions. This is a very useful method forresolving a perplexing retrieval problem of estimating soil roughness from SAR backscat-tering (Lee 2016). Verhoest et al. (2007) resolved uncertainty in soil roughness with aprobabilistic approach for the estimation of soil moisture with the European RemoteSensing (ERS) scatterometer. After generating the PDFs of soil moisture with backscatter-ing coefficients modeled with a possible range of Root Mean Square (RMS) height andcorrelation length, the mean value of the PDFs was used for determining optimal soilmoisture values. RMS errors in soil moisture was less than 6%. Notarnicola et al. (2006)attempted to find the optimal soil moisture in a vegetated area by taking an average ofPDFs. They assumed a Gaussian distribution for generating the PDFs of backscatteringcoefficient with an IEM simulation in various soil and vegetation water content condi-tions, and assessed the PDFs with soil parameter measurements. Synergy with optical dataimproved their parameterization of PDF shape.

In this study, we inverted time-varying soil roughness from backscattering to considerthe dynamics of high-resolution land surface heterogeneity. The objective of this study isto retrieve low to medium levels of soil moisture in sparsely vegetated soils without rely-ing on any soil parameter measurements in the field or multi-angular acquisitions, butwith an inversion method applicable to small scale-roughness.

GEOCARTO INTERNATIONAL 3

2. Methods

2.1. Study area

The Yanco site was established for the purpose of validating Soil Moisture Active Passive(SMAP) soil moisture data. It is a large flat area where barley and wheat are often grown.Two local stations at Yanco sites are used; Yanco-A3 station is (hereafter called YA3) andYanco-A5 (hereafter, YA5). We have chosen these sites, as they are considered to havestable roughness during specified periods. The YA3 local point station is located at�34.73521 latitude, and 146.08197 longitude (elevation: 132m) while local point stationYA5 is located at �34.712858 latitude and 146.127712 longitude (elevation: 136m). Thespatial domains are set to include these local point stations: latitude of �34.7399 to�34.7320 and longitude of 146.0780 to 146.0829 for YA3, and latitude of �34.7150 to�34.7110 and longitude of 146.1260 to 146.1299 for YA5. This corresponds to 2,475 pix-els (i.e., 45� 55 in latitude� longitude) for the YA5 site, while it is 4,895 pixels (i.e.,89� 55 in the same dimension) for the YA3 site. The spatial domain sizes do not includelarge-scale topography due to scale-dependency of roughness.

Soil texture for both sites has been found to be silty clay (conversation with Dr. JasonBeringer). During this experiment period, two sites were almost bare soils with no vegeta-tion growth. In details, the time-average of Leaf Area Index (LAI) was 0.225 at the YA3site, and 0.279 at the YA5 site, with respect to the maximum LAI value of 10. Soil moisturedata was acquired from the OzNet data archive (http://www.oznet.org.au/) (Smith et al.2012). It was measured with SDI-12 soil moisture probes every 20minutes. In this study,the 2018 soil moisture data of the top 5 cm layer was collected from Day of Year (DoY) 25to 145 at the YA3 site, and from DoY 1 to 145 in YA5 site are used. The soil has neverbeen frozen in these sites. The time-average of soil temperature at YA3 site is 22 �C duringexperiment period, while it is 27 �C at YA5 site. Unfortunately, the measurements of surfaceroughness (e.g., RMS height or correlation length) or rainfall data are unavailable.

2.2. Sentinel-1 data

Time-series of SAR backscattering data at C-land (5.4GHz) for the period of January toMay 2018 were used in this study. High resolution Interferometric Wide (IW) swathmode at Level-1 Ground Range Detected (GRD) has a spatial resolution of 20� 22m, anda pixel spacing of 10� 10m. Revisit time was 12 days over the Yanco sites. Polarization isVVþVH. Pass direction is a descending orbit with similar incidence angles of 35� 38�

over the study area.Following pre-processing procedure yielded gridded time-series SAR images stacked

over the experiment site and period. Because the VV-polarized IW data in GRD formatwas already multi-looked and projected to ground range at a pixel spacing of 10m (highresolution), radiometric and terrain correction (Range-Doppler Terrain Correction) werecarried out with ESA’s Sentinel Application Platform (SNAP)-python tool to mitigate geo-metric distortions in SAR data arising from topography or geometry, after calibrating itby Laur et al. (2004). After co-registration, multi-temporal speckle filtering was performedfor the time-series data sets.

2.3. SAR retrieval process

In this section, an inversion for surface roughness and soil moisture is described. Simply,it minimizes a difference between the Sentinel-1 backscattering measurement and IEM

4 J. H. LEE AND J. WALKER

forward model estimate. In other words, this algorithm retrieves the physical conditions(i.e., RMS height, correlation length, and dielectric constant) that IEM-modelled backscat-tering produces Sentinel-1 backscattering.

2.3.1. IEM backscattering modelThe IEM is a standard model to simulate the single backscattering coefficient under thephysical conditions defined by dielectric constant, surface roughness, and SAR configur-ation (Fung 1994). It is based upon an exponential relationship between backscatteringcoefficient and soil dielectric constant (Bindlish and Barros 2000). Its limitation is that thesensitivity of backscattering to soil moisture changes, depending on soils. In other words,it tends to be constant in wet soils, while it overestimates dry soils (Verhoest et al. 2007).

This IEM forward model computes the backscattering coefficients as follows (Ulabyet al. 1982; Shi et al. 1997):

ropp ¼k2

2expð�2k2

zs2ÞX1

n¼1

s2njInppj2Wnð�2kX; 0Þ

n!(1)

where subscript pp indicates the polarization, k is the wave number and S is the RMSheight. In addition, kx ¼ k sinh, kz ¼ k cosh, where, h is incidence angle. Ipp

n is a func-tion of the Fresnel reflection coefficient, relative permittivity and permeability of the sur-face (for a more detailed formula, see Chen et al. (1995); Shi et al. (1997)), and Wn is theFourier transform for the n th power of the normalized surface correlation function.

As seen from Equation (1), soil roughness and dielectric constant information arerequired to estimate backscattering with the IEM, in addition to SAR configuration suchas frequency or incidence angle. Such variables become dimensions of the LUT. In details,for the generation of LUT, RMS height ranged from 0.22 to 3.92 cm by an increment of0.05 cm, while correlation length ranged from 0.2 to 11.2 cm by an increment of 0.1 cm.The Autocorrelation Function (ACF, the height probability distribution function of ran-dom surface) has been set as an exponential shape for relatively smooth surface(Davidson et al. 2000). The real part of the dielectric constant starts from 2.93 to 52.92(Dobson et al. 1985; Reynolds et al. 2000; van der Velde et al. 2012).

2.3.2. Inversion of roughnessFrom the LUT generated in section 2.3.1, the retrieval algorithm selects soil roughnessconditions (i.e., RMS height and correlation length) to minimize a difference in backscat-tering between IEM simulation and Sentinel-1 measurement, as follows:

min jrsentinel�1 �rIEMðe, RHÞj (2)

where r is backscattering, and subscript describes the source of backscattering. IEM back-scattering is shown as a function of e dielectric constant and RH soil roughness.

Roughness was updated approximately monthly, due to high computational cost. Soilroughness is assumed to be invariant during three Sentinel-1 measurement data acquisi-tions. This is a reasonable assumption, as the site remained as bare soils with no dramaticchange in vegetation growth (see Supplementary Figure 1 for LAI) during the experimen-tal period. In addition, relatively smooth surface roughness (e.g., kS less than 3) wasassumed to ensure compliance with the IEM validity range (Mancini et al. 1999).

GEOCARTO INTERNATIONAL 5

2.3.3. Estimation of soil moistureWith surface roughness estimated in section 2.3.2., soil moisture is retrieved. The dielec-tric constant value that yielded the smallest difference between IEM backscattering simu-lated for the roughness condition estimated in 2.3.2. and Sentinel-1 backscattering isselected, as follows:

min jrsentinel�1 �rIEMðeÞj (3)

The limitation in this approach is that it assumes a stable surface condition. If the landsurface changes over a short period of time, as in the agricultural season, a temporal sam-pling in the current data acquisition scheme may be too slow to appropriately retrievethe roughness.

3. Results and discussions

3.1. Local point validation for time-series SAR soil moisture

YA3 site is consistently dry at 0.049m3/m3 on the time-average of soil moisture measuredin the field during experimental period. SAR retrievals also report such soil dryness at0.0372m3/m3. Retrieval accuracy is considered satisfactory. As shown in Table 1, the RootMean Square Error (RMSE) was found to be 0.01m3/m3, showing a marginal differencebetween IEM simulated and Sentinel-1 measured backscattering coefficients at 0.78 dB.The dry bias was also very small at � 0.0009m3/m3.

Time series soil moisture dynamics (top 5 cm) data for the YA3 site are presented inFigure 1. Some overestimation during DoY 100 to 120 is considered to be made because

Table 1. Time-average soil moisture data.

YA3 YA5

Average (m3/m3) 0.0372 0.1018RMSE (m3/m3) 0.0103 0.030033Bias (m3/m3) �0.0009 �0.015691Difference (dB)� 0.7808 0.012560�difference is the difference between IEM and Sentinel-1 backscattering.

Figure 1. Surface conditions over YA3 site in 2018: (a) surface roughness (b) validation of SAR soil moisture

6 J. H. LEE AND J. WALKER

actual soils were drier than the minimum levels assumed in LUT. Overall, this retrievalalgorithm appropriately estimates dry soils at YA3 site.

Unlike the YA3 site, time series soil moisture measurements were moderately wet inthe YA5 site (see Figure 2). As shown in Table 1, the time-average of SAR soil moistureis also moderate at 0.1m3/m3. RMSE was found to be 0.03m3/m3, showing satisfactoryresults for the difference between IEM simulated and Sentinel-1 measured backscatteringcoefficients at 0.0126 dB. Negative biases were found being � 0.016m3/m3. It seems to berelated to inversion errors in roughness (Lee 2016).

Based upon two different retrieval error structures discussed above, it is suggested that soilmoisture retrieval errors are related to dependency on physical variables such as soil moisture

Figure 2. Surface conditions over YA5 site in 2018: (a) surface roughness (b) validation of SAR soil moisture.

Figure 3. Surface roughness [cm] in YA3 site on DoY 60: (a) RMS height (b) correlation length.

GEOCARTO INTERNATIONAL 7

levels or surface roughness rather than season-dependent errors. However, this marginal levelof retrieval errors seems acceptable, suggesting that a monthly update of roughness is reason-able under current surface conditions (i.e., LAI of 0.1� 0.3). Thus, it is stated that despite asparse temporal sampling this retrieval method is appropriate when there is no weekly ordaily change in vegetation growth or surface roughness. This is considered as the improve-ment of Sentinel-1 data, as compared to ENVISAT ASAR data (van der Velde et al. 2012).

3.2. Spatial distribution of SAR soil moisture

After validating the retrieval method over the local point measurement data, it is appliedfor estimating the spatial distribution of soil moisture.

Figure 4. Spatial distribution for YA3 site on DoY 60: (a) soil moisture (b) difference in backscattering between IEMand Sentinel-1 and (c) Sentinel-1 backscattering. Units are indicated by color bars.

8 J. H. LEE AND J. WALKER

3.2.1. YA3 siteFigure 3 shows surface roughness over the area surrounding the YA3 site on DoY 60. InFigure 3(a), RMS height is shown to be spatially heterogeneous, ranging from 0.5 to1.5 cm. The spatial average of RMS height was found to be 0.99 cm. In Figure 3(b), simi-larly to RMS height, the correlation length was also found to be spatially heterogeneous,ranging from 3.0 to 4.0 cm. The spatial average of correlation length was found to be3.33 cm. The surface soil moisture retrieved with this roughness is further shown inFigure 4. In Figure 4(a), surface soil moisture was spatially mostly dry. The spatial averagewas 0.052m3/m3.This is in agreement with the low soil moisture estimation of local fieldmeasurements shown in Figure 1.

It is thus stated that a spatial distribution of backscattering in YA3 site was related toroughness. More specifically, areas with high backscattering (shown yellow in Figure 4(c))were closely related to RMS height in Figure 3(a) and low soil moisture in Figure 4(a). Inother words, high backscattering is considered mostly due to roughness rather than soilmoisture. The retrieval for these dry soils has a reasonable certainty, suggesting a differ-ence between IEM simulated and Sentinel-1 measured backscattering was less than0.89 dB. Thus, uncertainty is mostly low, as shown by dark blue in Figure 4(b). In con-trast, other areas higher than 4 dB of backscattering difference (shown yellow in Figure4(b)) suggest that a retrieval algorithm might underestimate (or overestimate) soil mois-ture there. As described in Section 2.3.1. and 3.1., error sources may be related to anassumption converting dielectric constant to soil moisture. It may be also affected bysome ground features, or geometric errors in SAR (Zhu et al. 2019).

3.2.2. YA5 siteFigure 5 shows surface roughness over the YA5 site on DoY 121. Unlike the YA3 site,RMS height was spatially random, ranging from 0.52 to 2 cm in Figure 5(a). The spatialaverage of RMS height was found to be 0.796 cm. In Figure 5(b), similarly to RMS height,the correlation length was also random. The spatial average of correlation length wasfound to be 3.1 cm.

Surface soil moisture retrieved with this roughness condition is shown in Figure 6. InFigure 6(a), surface soil moisture was found to be moderately wet. Quantitatively, the spatialaverage of soil moisture was 0.1135m3/m3. This level of soil moisture is similar to that foundfrom local field measurements in Figure 2 and Table 1. Uncertainty for this estimation was

Figure 5. Surface roughness [cm] in YA5 site on DoY 121: (a) RMS height and (b) correlation length.

GEOCARTO INTERNATIONAL 9

also very little. The spatially averaged difference between the IEM simulated and Sentinel-1measured backscattering was found to be very small at 0.024 dB in Figure 6(b).

It is suggested that the area with high backscattering (shown yellow in Figure 6(c)) wasnot directly related to soil moisture in Figure 6(a) or RMS height in Figure 5(a).Considering that overall uncertainty in retrieval is very low in Figure 6(b), it is stated thatseveral factors other than soil moisture or roughness can elevate backscattering in a com-plicated way, and it is inappropriate to establish a direct and linear relationship betweensoil moisture and backscattering.

4. Conclusion

ENVISAT ASAR was the first SAR sensor that produces a global coverage over the landsurface. However, for some operational reasons, a spatial coverage was irregular, and atemporal resolution was low. Thus, it was difficult to collect time-series data. AlthoughASCAT was operationally used for weather prediction models, its spatial resolution wasstill low. As compared to ASAR or ASCAT, Sentinel-1 data has improvements in terms ofspatio-temporal resolutions. This study explored a retrieval algorithm to handle it in acomputationally effective way.

Figure 6. Spatial distribution in YA5 site on DoY 121: (a) soil moisture (b) difference in backscattering between IEMand Sentine-1 (c) original Sentinel-1 backscattering. Units are indicated by color bars.

10 J. H. LEE AND J. WALKER

For Sentinel-1 data, this study found that a monthly retrieval of surface roughnessfrom time series Sentinel-1 backscattering measurements is sufficient for soil moistureretrievals, if surface condition is stable at LAI of 0.1� 0.3. In detail, a high spatial vari-ability of surface roughness at this small scale is appropriately expressed by this inversionapproach, as spatial differences of backscattering between IEM simulated and Sentinel-1were found to be small at 0.78 dB in YA3 site and 0.01 dB in YA5 site, respectively. theYA3 site in dry soils exhibited spatially patterned roughness, which was considered as themain contributor to backscattering variation. In contrast to the YA3 site, backscatteringintensity in the YA5 site was found to have a more complicated relationship with soilmoisture. Thus, it is suggested that this method fairly disentangled soil moisture from thebackscattering effects of other physical parameters such as roughness, and high backscat-tering cannot be directly interpreted as wet soils. .

Due to assumption used for a retrieval of roughness, a change in vegetation conditionsor surface roughness should be well-monitored during a retrieval of roughness with thisalgorithm. For example, a satellite-retrieved vegetation index should be associated withthis retrieval in order to monitor any change in land surface. In addition, large scale top-ography would not be included for study domain. Thus, future works may suggest a strat-egy to invert roughness and soil moisture from SAR backscattering under the effect ofvolumetric scattering arising from time-varying vegetation. In addition, they will alsoinvestigate whether artificial neutral network can effectively estimate roughness despitethis complicate and non-linear relationship between soil moisture and backscattering.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Funding

This work was supported by the National Research Foundation of the Korean government (NRF-2018R1D1A1B07048817).

References

Bauer-Marschallinger B, Freeman V, Cao S, Paulik C, Schaufler S, Stachl T, Modanesi S, Massari C,Ciabatta L, Brocca L, et al. 2019. Toward global soil moisture monitoring with Sentinel-1: harnessingassets and overcoming obstacles. IEEE Trans Geosci Remote Sens. 57(1):520–539.

Beven K, Freer J. 2001. Equifinality, data assimilation, and uncertainty estimation in mechanistic model-ling of complex environmental systems using the GLUE methodology. J Hydrol. 249(1–4):11–29. doi:.(01)00421-8

Bindlish R, Barros AP. 2000. Multifrequency soil moisture inversion from SAR measurements with theuse of IEM. Remote Sens Environ. 71(1):67–88. doi:. (99)00065-6

Chen KS, Yen SK, Huang WP. 1995. A simple model for retrieving bare soil moisture from radar-scattering coefficients. Remote Sens Environ. 54(2):121–126. doi:. (95)00129-O

Davidson MWJ, Thuy Le T, Mattia F, Satalino G, Manninen T, Borgeaud M. 2000. On the characteriza-tion of agricultural soil roughness for radar remote sensing studies. IEEE Trans Geosci Remote Sens.38(2):630–640.

Dobson MC, Ulaby FT, Hallikainen MT, El-Rayes MA. 1985. Microwave dielectric behavior of wet soil-Part II: dielectric mixing models. IEEE Trans Geosci Remote Sens. GE-23(1):35–46.

Fung AK. 1994. Microwave scattering and emission models and their applications. Norwood, MA: ArtechHouse.

GEOCARTO INTERNATIONAL 11

Henry JB, Chastanet P, Fellah K, Desnos YL. 2006. Envisat multi-polarized ASAR data for flood mapping.Int J Remote Sens. 27(10):1921–1929.

Jensen D, Reager JT, Zajic B, Rousseau N, Rodell M, Hinkley E. 2018. The sensitivity of US wildfireoccurrence to pre-season soil moisture conditions across ecosystems. Environ Res Lett. 13(1):014021.

Kerr YH, Waldteufel P, Richaume P, Wigneron JP, Ferrazzoli P, Mahmoodi A, Al Bitar A, Cabot F,Gruhier C, Juglea SE, et al. 2012. The SMOS soil moisture retrieval algorithm. IEEE Trans GeosciRemote Sens. 50(5):1384–1403.

Kornelsen K, Coulibaly P. 2013. Advances in soil moisture retrieval from synthetic aperture radar andhydrological applications. J Hydrol. 476:460–489.

Laur H, Bally P, Meadows P, Sanchez J, Schaettler B, Lopinto E, Esteban D. 2004. Derivation of the back-scattering coefficient in ESA ERS SAR PRI products. ESA/ESRIN ES-TN-RS-PM-HL09, 2: 1–55.

Lee HJ. 2016. Sequential ensembles tolerant to synthetic aperture radar (SAR) soil moisture retrievalerrors. Geosciences. 6(2):19.

Lee JH, Pellarin T, Kerr YH. 2014. Inversion of soil hydraulic properties from the DEnKF analysis ofSMOS soil moisture over West Africa. Agric for Meteorol. 188:76–88. doi:.

Mancini M, Hoeben R, Troch PA. 1999. Multifrequency radar observations of bare surface soil moisturecontent: A laboratory experiment. Water Resour Res. 35(6):1827–1838.

Merzouki A, McNairn H, Pacheco A. 2011. Mapping soil moisture using RADARSAT-2 data and localautocorrelation statistics. IEEE J Sel Top Appl Earth Obs Remote Sens. 4(1):128–137.

Mirsoleimani HR, Sahebi MR, Baghdadi N, El Hajj M. 2019. Bare soil surface moisture retrieval fromSentinel-1 SAR data based on the calibrated IEM and Dubois models using neural networks. Sensors.19(14):3209.

Niu G-Y, Yang Z-L, Mitchell KE, Chen F, Ek MB, Barlage M, Kumar A, Manning K, Niyogi D, Rosero E,et al. 2011. The community Noah land surface model with multiparameterization options (Noah-MP):1. Model description and evaluation with local-scale measurements. J Geophys Res. 116:D12109.doi:10.1029/2010JD015139.

Notarnicola C, Angiulli M., Posa F. 2006. Use of radar and optical remotely sensed data for soil moistureretrieval over vegetated areas. IEEE Trans Geosci Remote Sens. 44, 925–935.

Paloscia S, Pettinato S, Santi E, Notarnicola C, Pasolli L, Reppucci A. 2013. Soil moisture mapping usingSentinel-1 images: algorithm and preliminary validation. Remote Sens Environ. 134:234–248. doi:.

Pauwels VRN, Hoeben R, Verhoest NEC, De Troch FP, Troch PA. 2002. Improvement of TOPLATS-based discharge predictions through assimilation of ERS-based remotely sensed soil moisture values.Hydrol Process. 16(5):995–1013.

Pierdicca N, Pulvirenti L, Bignami C. 2010. Soil moisture estimation over vegetated terrains using multi-temporal remote sensing data. Remote Sens Environ. 114(2):440–448. doi:.

Rahman MM, Moran MS, Thoma DP, Bryant R, Sano EE, Holifield Collins CD, Skirvin S, Kershner C,Orr BJ. 2007. A derivation of roughness correlation length for parameterizing radar backscatter models.Int J Remote Sens. 28(18):3995–4012.

Reynolds CA, Jackson TJ, Rawls WJ. 2000. Estimating soil water-holding capacities by linking the Foodand Agriculture Organization Soil map of the world with global pedon databases and continuous pedo-transfer functions. Water Resour Res. 36(12):3653–3662.

Shi JC, Wang J, Hsu AY, Neill PEO, Engman ET. 1997. Estimation of bare surface soil moisture and sur-face roughness parameter using L-band SAR image data. IEEE Trans Geosci Remote Sens. 35(5):1254–1266.

Smith AB, Walker JP, Western AW, Young RI, Ellett KM, Pipunic RC, Grayson RB, Siriwardena L,Chiew FHS, Richter H. 2012. The Murrumbidgee soil moisture monitoring network data set. WaterResour Res. 48(7):W07701. doi:10.1029/2012WR011976.

Taylor CM, de Jeu RAM, Guichard F, Harris PP, Dorigo WA. 2012. Afternoon rain more likely over driersoils. Nature. 489(7416):423–426.

Ulaby F, Moore R, Fung A. 1982. Microwave remote sensing: Active and passive. In: Ulaby F, editor.Radar remote sensing and surface scattering and emission theory (Vol. 2). Norwood, MA: ArtechHouse; p. 484–487.

van der Velde R, Su Z, van Oevelen P, Wen J, Ma Y, Salama MS. 2012. Soil moisture mapping over thecentral part of the Tibetan Plateau using a series of ASAR WS images. Remote Sens Environ. 120:175–187.

Verhoest NEC, De Baets B, Mattia F, Satalino G, Lucau C, Defourny P. 2007. A possibilistic approach tosoil moisture retrieval from ERS synthetic aperture radar backscattering under soil roughness uncer-tainty. Water Resour Res. 43(7):W07435. doi:10.1029/2006WR005295.

12 J. H. LEE AND J. WALKER

Wagner W, Noll J, Borgeaud M, Rott H. 1999. Monitoring soil moisture over the Canadian Prairies withthe ERS scatterometer. IEEE Trans Geosci Remote Sens. 37(1):206–216.

Wagner W, Sabel D, Doubkova M, Bartsch A, Pathe C. 2010. The potential of Sentinel-1 for monitoringsoil moisture with a high spatial resolution at global scale. In: Earth Observation and Water CycleScience, Frascati, Italy, Volume: ESA SP-674, November 2010.

Zhu L, Walker JP, Ye N, R€udiger C. 2019. Roughness and vegetation change detection: A pre-processingfor soil moisture retrieval from multi-temporal SAR imagery. Remote Sens Environ. 225:93–106.

GEOCARTO INTERNATIONAL 13